Strategies for temperature-compensating differential pairs

l

Your really need to explain your circuit better. I have a gut feeling you don't mean a differential pair since the collectors would be high impedance and just form gain stages if fed a reference current.

You've lost me already. What do you mean by putting a current into the collector of a diff pair? Normally one stuffs a current into the paralleled emitters.

Can you post a schematic?

John

and then take an output current from the collector of

Sorry for not being more clear - in the meantime I was able to find a website that has an example of the circuit configuration I am talking about here, figure 3:

It appears from the text the most common way to compensate the 1/T temperature dependence of the circuit is to use a resistor with a positive temperature coefficient in the circuit that provides the control voltage. To make it work well I guess you would want to bind the differential pair to the tempco resistor with some kind of thermally conductive adhesive.

ns

will

d

Looks like frames did you a disservice. It this the page in question?

If not, then right click on the link in question and open it in another window to get the correct URL.

Thanks, that's the correct page.

What _exactly_ are you trying to do... compensate a diff pair gain over temperature? ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
      The only thing bipartisan in this country is hypocrisy

Hi Jim - I'm interested in learning more about what methods work best for compensating the 1/T temperature dependence in the current source circuit shown about halfway down the page here:

The simplest method is apparently to use a voltage divider that contains a resistor with a positive temperature coefficient, but it's not entirely accurate and only works over a limited temperature range. Other methods are mentioned (keeping the circuit in a heater, using a multiplier) but I'd like to know what sort of method the experienced analog designers here would prefer.

That's where I'm lost. I see no redeeming social value for the circuit. What will be the ultimate end use? The example seems to be from someone with less than a full understanding of bipolar transistors... although maybe he's grasping at making a log amplifier... and attacking it cattywampus??

I posted a chip circuit here some time ago, can't seem to find it now (if anyone remembers, tell me what filename I posted it as :-), but it's gain in dB's was linear with a control voltage. All temperature compensation on chip, no thermistors. ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
      The only thing bipartisan in this country is hypocrisy

the

a g

ind

ally

a

ations

Vbe will

n of

the

used

=A0 =A0 ...Jim Thompson

Don't bother with that website. With a bandgap, opamp, small signal transistor (BJT or fet), and resistor, you can make a constant current source. The bandgap gives you the constant voltage. You can probably find a thousand V to I circuits on line, but basically you have an op amp output feeding the gate/base of a small signal transistor. The source/emitter connects to a resistor. The other end of the resistor goes to ground. The voltage on the intersection of the small signal transistor and resistor goes back to the op amp negative input. The op amp positive input goes to the voltage reference. Thus you end up forcing the voltage reference voltage to appear across the resistor. I=3DV/R. The current is sinked from the collector/drain of the small signal device. Clearly if you use a bipolar, there is going to be an error due to beta, and a tempco since beta is a function of temperature. A mosfet would be very accurate since there is no error due to base current, but probably overkill.

The small signal device is a follower, but even a follower has phase shift, so you want to use a relatively overcompensated op amp in the application.

There is a bipolar current source design called the "peaking" reference. It is not really very good, but might be worth the read if you can dig it up. I've used it in non-critical designs.

In MOS, you can find a sweet spot when the effect of the threshold tempco and the mobility tempco compensate each other. That is used as another cheap ass current reference, but of course nothing is better than a bandgap and V to I.

I can't tell what he's trying to accomplish.

Don't bother with that website. With a bandgap, opamp, small signal transistor (BJT or fet), and resistor, you can make a constant current source. The bandgap gives you the constant voltage. You can probably find a thousand V to I circuits on line, but basically you have an op amp output feeding the gate/base of a small signal transistor. The source/emitter connects to a resistor. The other end of the resistor goes to ground. The voltage on the intersection of the small signal transistor and resistor goes back to the op amp negative input. The op amp positive input goes to the voltage reference. Thus you end up forcing the voltage reference voltage to appear across the resistor. I=V/R. The current is sinked from the collector/drain of the small signal device. Clearly if you use a bipolar, there is going to be an error due to beta, and a tempco since beta is a function of temperature. A mosfet would be very accurate since there is no error due to base current, but probably overkill.

The small signal device is a follower, but even a follower has phase shift, so you want to use a relatively overcompensated op amp in the application.

There is a bipolar current source design called the "peaking" reference. It is not really very good, but might be worth the read if you can dig it up. I've used it in non-critical designs.

In MOS, you can find a sweet spot when the effect of the threshold tempco and the mobility tempco compensate each other. That is used as another cheap ass current reference, but of course nothing is better than a bandgap and V to I.

Here's one trick I've seen described in IC design where you need stable reference voltages at many sites all around a large chip. They don't want to generate a voltage and send it along a long trace with significant resistance because of voltage drop, so they generate a current and mirror out multiple copies of it, and send currents anywhere on the chip they need a reference. Like this: p-channel mosfets Vdd | ,-------+-------, | | | |\ _| _| _| Vbg-------|-\ | | | | >---||------||------|| ,--|+/ |_ |_ |_ | |/ | | | | | | | '------------+ Iout '-------+--Vbg | Iout | | | R R | | | | gnd gnd

The op amp maintains whatever current through the resistor is necessary to keep it at the bandgap voltage. How much current it makes will depend on the tempco of the resistor. The mirrored current gets to its destination and is fed into a resistor of the same composition; the resistor tempco cancels.

That guy is such a bad writer. He never gives you the most basic information: what's the circuit supposed to do, what kind of input signal goes in, just the most basic stuff that should be the first thing he tells you. Instead he starts out from nowhere and goes into excructiating detail about a whole bunch of stuff that has no context. It looks like he's describing an attempt to build a transconductance amplifier with an exponential transfer function. And there's a sentence buried deep in the text, something about in input of one volt per octave, and he wants the output current to double for a change of one volt in the input signal. I haven't figured out whether the circuit he shows is fit for that purpose. How many octaves (volts) is the input going to vary? How big of a signal can you put into the circuit he drew? You could spend forever getting lost in the details. Better to ask first: what's the right approach?

Indeed, it is a form of a log amplifier. In the analog synthesizer world (pre-MIDI) there was a standard that control voltages feeding the modules should cause a 1 octave change in oscillator frequency, filter cutoff frequency, etc. for each 1 volt change in input. So if you take the output of that current source circuit and put the current into an integrator you'll get a ramp out, the frequency of which is directly proportional to the current input. Connect up some kind of circuit to discharge the integrator capacitor when it hits some threshold and you get a sawtooth wave.

To get the 1 volt per octave curve you solve the equation VT ln (2) and you find delta Vbe is something like 17 millivolts, so there's an input control summer that divides down the input control voltage to that level and is adjustable with trimmer resistors. The differential structure cancels out the first order temperature dependence of Is, but it will still drift around with the dependence of VT on T. I've found that apparently the way this problem was typically "solved" was a thermistor was used in the feedback path of the control summer, but it never really worked perfectly.

Once microprocessors started being used in analog synthesizers several of the more expensive models had "tune" buttons where the microprocessor would do its best to dynamically apply a control voltage that would get the oscillators in tune. Soon synthesis went digital and they just became computers in a fancy box. Interestingly, though, Korg recently came out with a true analog synthesizer for the first time in decades - though its musical usefulness might be questioned:

In any case, the circuit as I've described it above is probably about 40 years old. I thought that there must certainly be a better way to do things now from the analog perspective?

It does make more sense when you get to the page above through the main page:

the material then has some context!

the

a g

ind

ally

a

Vbe

n of

the

ignal

lls

detail

the

=A0I

e.

gnal

g lost

I thinks frames messed you up again. I have no idea where you are trying to send us. However, I don't think that website is particularly useful. Not only is the electronics questionable, but the page refers to figures that are not even labeled.

SSM and Curtis Electro Music used to make chips for analog synthesizers. ADI absorbed SSM. Doug Curtis died a few years ago, and he had already folded that company into his design service company. THAT makes many of the same analog circuits. I would try to find designs from those companies rather than mess with that questionable website.

Don't get me wrong. I would like to encourage you to do the analog design since nowadays people think engineering is programming a damn pic. Just don't put too much faith into one website. This stuff really isn't rocket science, but analog can be difficult to understand if it is not explained well.

OK, so the problem statement is "Output current doubles for each one volt increase in input... temperature compensated" ?? ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
      The only thing bipartisan in this country is hypocrisy

More succinctly, yes that's about it. The first part is not difficult, but the second part was/is apparently a bigger challenge, as there are plenty of old synthesizers that could never stay well in tune.

• HUH??? _no_ figures; where shall one look???

So, use a THAT2181 as your gain stage, and attenuate the gain-control signal with a two-resistor attenuator. R1 is a straight resistor, say

40k ohms, and R2 is a 1kohm +3300ppm/C PTC thermistor. The thermistor has to be thermally coupled to the amplifier IC, of course. The 'Ec' program pins were made for this kind of thing.